Low-Cost Descriptors of Electrostatic and Electronic Contributions to OPEN ACCESS Anion Redox Activity in Batteries

IOP SciNotes 1 (2020) 024805 https://doi.org/10.1088/2633-1357/ab9750

ARTICLE Low-cost descriptors of electrostatic and electronic contributions to OPEN ACCESS anion redox activity in batteries

RECEIVED 19 April 2020 Daniel W Davies1,2 , Benjamin J Morgan2,3 , David O Scanlon2,4,5 and Aron Walsh1,2,6 REVISED 1 Department of Materials, Imperial College London, Exhibition Road, London SW7 2AZ, United Kingdom 18 May 2020 2 The Faraday Institution, Quad One, Harwell Science and Innovation Campus, Didcot, United Kingdom ACCEPTED FOR PUBLICATION 3 Department of Chemistry, University of Bath, Claverton Down, Bath BA2 7AY, United Kingdom 28 May 2020 4 University College London, Department of Chemistry and Thomas Young Centre, 20 Gordon Street, London WC1H 0AJ, United PUBLISHED Kingdom 21 July 2020 5 Diamond Light Source Ltd., Diamond House, Harwell Science and Innovation Campus, Didcot, Oxfordshire OX11 0DE, United Kingdom Original content from this 6 Department of Materials Science and Engineering, Yonsei University, Seoul 03722, Republic of Korea work may be used under the terms of the Creative E-mail: [email protected] Commons Attribution 4.0 licence. Keywords: anion redox, cathodes, descriptors, computational screening, python Any further distribution of this work must maintain attribution to the author(s) and the title of Abstract the work, journal citation and DOI. Conventional battery cathodes are limited by the redox capacity of the transition metal components. For example, the delithiation of LiCoO2 involves the formal oxidation from Co(III) to Co(IV). Enhanced capacities can be achieved if the anion also contributes to reversible oxidation. The origins of redox activity in crystals are difficult to quantify from experimental measurements or first- principles materials modelling. We present practical procedures to describe the electrostatic (Madelung potential) and electronic (integrated density of states) contributions, which are applied to the LiMO2 and Li2MO3 (M=Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zr, Nb, Mo, Ru, Rh, Pd, Ag, Hf, Ta, W, Re, Os, Ir, Pt, Au) model systems. We discuss how such descriptors could be integrated in a materials design workflow.

The large difference in electronegativity between metals and oxygen mean that metal oxides are heteropolar compounds [1, 2]. Oxidation of metal oxides occur between two extremes. In the ﬁrst, common for transition metal oxides, an increase in the formal oxidation state of the metal occurs (MMnn 1). In the second, the oxide itself is oxidised (OO21 ), resulting in a so-called oxygen hole [3]. Due to the electron conﬁguration of the oxide ion going from closed-shell 2p6 to open-shell 2p5, evidence for oxygen hole formation has commonly being reported from electron paramagnetic resonance (EPR) spectroscopy. As early as 1955, O’Brien and Bleaney used EPR to establish that the colour centres in smokey

quartz (SiO2) are due to oxygen redox induced by Al impurities [4]. EPR signals are presented as clear evidence for the hopping of holes between oxygen sites in SiO2 in the later work of Schnadt and Schneider [5]. The dominant oxidation mechanism in conventional intercalation battery cathodes is cationic, while recently there has been a strong focus on the development of so-called anion redox cathodes. In this context, anion redox refers to oxidation processes in which M and O are both active [6], which could enhance overall battery capacity. Detailed reviews on this topic can be found at [7] and [8]. The driving forces dictating to what extent M and O can be oxidised in the same compound are challenging to probe by experiment and simulation. Recently, there have been substantial efforts made to understand how anion redox might be harnessed in a small set of emerging cathode materials [9–16]. Meanwhile, developments in predictive, high-throughput design in other areas of energy materials research such as photovoltaics [17–19], photocatalysis [20, 21] and thermoelectrics [22–24] have enabled broader searches of the chemical landscape and have in some cases led to the discovery of promising candidate compounds. Such virtual screening studies rely on the use of simple descriptors to predict properties of candidate materials at low computational cost.

Figure 1. (a) Average Madelung potential at oxygen sites in LiMO2 and Li2MO3 when the M transition metal is varied. Crystal structures of (b) LiMO2 and (c) Li2MO3 prototypes are shown with Li, M and O depicted as green, blue, and red spheres, respectively.

Here we present practical procedures for probing the contributions to redox phenomena from electrostatic and electronic perspectives. Our workﬂow relies on open-source materials informatics tools and the output of routine ﬁrst-principles calculations. We provide worked examples to demonstrate our approach, which is applied to LiMO2 and Li2MO3 model systems (M=Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zr, Nb, Mo, Ru, Rh, Pd, Ag, Hf, Ta, W, Re, Os, Ir, Pt, Au), that have been structurally optimized using density functional theory (DFT).

1. Electrostatic descriptors

Oxide (O2−) ions are unstable in vacuum, but become stable in a crystal due to the conﬁning electrostatic potential [25]. This makes the positioning of the O 2p valence band in a metal oxide highly sensitive to the crystal environment [26]. The electrostatic (Madelung) energy of a crystal under the ionic point-charge approximation can be zzij expressed as the sum of all pairwise interactions, E ij, , where zi and zj are the formal charges of the ith and rij jth ions and rij is the interionic distance. Accordingly, the potential experienced at site i (the site potential) is zj given as Vi j . A larger potential is produced for more highly charged metals at shorter distances. rij A smaller average Madelung site potential might be expected from a higher Li+:Mn+ ratio, owing to the overall decrease in positive charge. For stoichiometric systems, however, the value of n increases with this ratio in order to satisfy the valence of all species, and this is the dominant effect as seen in Figure 1 when comparing 3+ 4+ LiMO2 (M ) to Li2MO3 (M ) for a given transition metal. There is a range of >3.5 V for the average Madelung potential of oxygen even within a given structure prototype, (LiAuO2 vs LiCoO2, Figure 1(a)). Given that the formal charge of the transition metals remains constant within a series, the variation in Vi is solely to differences in interionic distances in the relaxed crystal structures. It should be noted that in addition to the compounds containing group 10 metals (Ni, Pd, Pt), which relax to a different structure type with square planar geometry, Nb, Ta and Os also distort the parent structure shown in Figure 1.

If Li/M site disorder is introduced—based on 128 atom 2×2×2 supercells of LiMO2—the impact on the local site potentials is large. This effect is apparent in Figure 2, which shows the oxide potential range in LiMO2 for the first five transition metals that we consider. The site potential decreases rapidly as the Li+:M3+ ratio increases in a similar way for all five compounds. This trend recovers the analysis of Seo et al, which showed that a Li-rich coordination is required for the extraction of oxygen electrons in Li-ion cathodes [11].

2 IOP SciNotes 1 (2020) 024805 D W Davies et al

O + 3+ Figure 2. Average Madelung potential at oxygen sites (VM ) in LiMO2 for different Li :M nearest neighbour ratios.

M ﬁ Figure 3. Hole localization energy (Eh) and the negative of the 4th ionization potential of M (In ) for LiMO2 where M is a rst-row transition metal (Ti–Cu).

Finally, we can extend this analysis to estimate whether oxidation will proceed via M or O. In the context of oxide superconductors, Torrance and Metzger deﬁned hole localisation energy as

OOM M EAIh ()()2 n eVVMM ()1 O M fi which separates into (i) atomic contributions A2 andIn , the second electron af nity of oxygen and the nth ionization energy of the M()n1 cation, respectively, and (ii) structural contributions from the Madelung O M potentials VM and VM [27]. A smaller value of Eh indicates more favourable hole formation on oxygen. M Figure 3 shows how this descriptor largely follows the same trend as the negative of In , but is modulated by the structural contributions via the Madelung potential. Note that this trend is most obviously observed for the first row transition metals (Figure 3) for which a complete set of reliable ionisation energies is available. In summary, the Madelung site potential is a physically meaningful metric that is simple to calculate. While there are few theoretical or experimental results with which the observations made on our model systems can be directly compared, we have shown how the O site potential can be highly sensitive to the surroundings in typical Li-ion cathode compounds, that the observed trends are consistent with recent theoretical work in the field of anion redox [11], and how this can be extended to estimate hole localisation energy.

2. Electronic descriptors

A quantitative description of cathode redox processes require explicit consideration of delithiation. In the spirit of ﬁnding computationally cheap descriptors, however, we can examine the lithiated material [6, 7]. The electronic density of states (DOS) represents the number of states available as a function of energy, and is a standard output of all electronic structure codes. We can project this onto atomic sites to obtain a species and orbital decomposed DOS. By integrating the oxidation window in the upper valence band, it is possible to extract

3 IOP SciNotes 1 (2020) 024805 D W Davies et al

Figure 4. Orbital contributions to electronic density of states (DOS) in MgO and Cu2O. The upper valence band (highlighted blue) in MgO is dominated by O p-states (100%) while in Cu2O it is dominated by Cu d-states (83%). Plots created with Sumo command-line tools [28].

Figure 5. Orbital contributions to the upper valence band (highlighted blue) of the electronic density of states (DOS) in LiRuO2 (20% O p, 80% Ru d), Li2RuO3 (35% O p, 65% Ru d), LiWO2 (18% O p, 82% W d) and Li2WO3 (25% O p, 75% W d). Plots created with Sumo command-line tools [28]. the ratio of metal d-orbital (M d) to oxygen p-orbital (Op) contributions. Here we choose an oxidation window of 1 eV below the top of the valence band, but this could be easily adapted to more accurate distributions. + 10 Figure 4 shows the extreme cases of MgO and Cu2O, in which Cu has a d configuration. DOS analysis (Notebook 2) shows that the upper valence band is comprised of 100% O p and 0% Mg s for MgO, and 87% Cu d and 13% O p for Cu2O. These fractions are expressed in terms of the orbitals of interest (e.g. Cu d + O p=100% for Cu2O) to avoid influence on the stoichiometry, which will change upon cycling. The DOS fractions of M-d and O-p are compared for LiRuO2, Li2RuO3, LiWO2 and Li2WO3 in Figure 5. The O-p contribution to the upper valence band is higher for Li2MO3 compared to LiMO2. It should also be noted that the ground state magnetic ordering of Li2RuO3 is ferromagnetic (Figure 5), which is necessary to obtain a reliable electronic structure. An efficient way to automate the search for magnetic ground states is described by Horton et alin [29]. This analysis allows for an immediate insight into which states are available in the upper valence band to contribute to redox processes. Our approach exemplifies how tracking individual orbital contributions to the DOS, which has constituted important evidence for the presence of absence of anionic redox in previous studies [12, 30], can be quantified.

3. Workﬂow integration

The electrostatic descriptors are low-cost and only require structural information. Calculation of Madelung site potentials can be carried out in a few seconds on a desktop computer using the Ewald method, which is implemented in Pymatgen [31] (see Notebook 1). While Madelung energies have previously been used as descriptors of battery voltage [32], individual site potentials have been applied to understanding band

4 IOP SciNotes 1 (2020) 024805 D W Davies et al

Figure 6. Workﬂow for a generalised Monte Carlo simulation where the Metropolis algorithm [35] is used to build up an acceptable set of system conﬁgurations. A step to calculate both the Madelung potential and density of states descriptors could be inserted (orange box) at negligible additional computational cost.

alignments [33, 34]. The approaches could be improved by incorporating dielectric polarisation of the host crystal. The electronic descriptors require only the DOS, and this is produced as standard output for DFT calculations. The post-processing is performed in less than a second on a desktop computer; a negligible amount of time compared to the original DFT calculations (see Notebook 2). Overall, both descriptors require common representations of crystalline compounds (i.e. crystal structure and electronic DOS) and are therefore suitable for large-scale studies of lithiation, sodiation in Na-ion battery cathodes, and other chemistries. They could both be integrated, for example, into Monte Carlo techniques as part of multi-scale simulations (Figure 6).

4. Technical details

All the code and data necessary to reproduce the results discussed here can be found the accompanying data repository at https://doi.org/10.5281/zenodo.3754712. We rely primarily on the Pymatgen Python library [31], which is used to process crystal structures optimized by DFT (Notebook 1) and electronic structures solved by hybrid DFT (Notebook 2). All ﬁrst-principles calculation were automated using the Python packages Atomate [36] and Fireworks [37]. We used Kohn–Sham DFT with a projector-augmented plane wave basis [38] as implemented in the Vienna Ab-initio Simulation Package (VASP) [39, 40]. We use the PBEsol exchange-correlation functional [41] and a

5 IOP SciNotes 1 (2020) 024805 D W Davies et al k-point grid is generated for each calculation with a density of 120 Å3 in the reciprocal lattice. The kinetic-energy cut-off is set at 600 eV and the forces on each atom minimised to below 0.005 eVÅ−1. Semi-local exchange-correlation treatments such as the PBEsol functional provide an accurate description of crystal structures but tend to underestimate the electronic bandgaps of semiconductors. To overcome this issue, more accurate calculations are performed using a hybrid non-local functional HSE06 [42], which includes 25% screened Hartree–Fock exact exchange, which are used for the electronic structure analysis. Γ-centred homogeneous k-point grids are used with a density of 120 Å3 in the reciprocal lattice and the kinetic energy cutoff is set at 550 eV.

Acknowledgments

This work was carried out with funding from the Faraday Institution (faraday.ac.uk; EP/S003053/1) grant number FIRG003 and FIRG017, and used the MICHAEL computing cluster. BJM is supported by a Royal Society University Research Fellowship (No. UF130329). Via our membership of the UK’s HEC Materials Chemistry Consortium, which is funded by EPSRC (EP/L000202), this work used the ARCHER UK National Supercomputing Service (http://www.archer.ac.uk).

Data availability statement

The data that support the ﬁndings of this study are openly available at 10.5281/zenodo.3754712.

ORCID iDs

Daniel W Davies https://orcid.org/0000-0003-4094-5992 Benjamin J Morgan https://orcid.org/0000-0002-3056-8233 David O Scanlon https://orcid.org/0000-0001-9174-8601 Aron Walsh https://orcid.org/0000-0001-5460-7033

References

[1] Mott N F and Gurney R W 1948 Electronic Processes in Ionic Crystals (Oxford: Clarendon) [2] Walsh A, Sokol A A, Buckeridge J, Scanlon D O and Catlow C R A 2018 Nature Mater. 17 958 [3] Possenriede E, Hellermann B and Schirmer O 1988 Sol. Stat. Commun. 65 31 [4] O’Brien M C M and Bleaney B 1955 Proc. of the Royal Society of London. Series A. Mathematical and Physical Sciences 231 404 [5] Schnadt R and Schneider J 1970 Phys. Kondens. Mater. 11 19–42 [6] Rouxel J 1997 Curr. Sci. 73 31 (https:/jstor.org/stable/24098143) [7] Assat G and Tarascon J-M 2018 Nature Energy 3 373 [8] Saubanère M, McCalla E, Tarascon J-M and Doublet M-L 2016 Energy Environ. Sci. 9 984 [9] House R A, Jin L, Maitra U, Tsuruta K, Somerville J W, Förstermann D P, Massel F, Duda L, Roberts M R and Bruce P G 2018 Energy & Environmental Science 11 926 [10] Maitra U et al 2018 Nat. Chem. 10 288 [11] Seo D-H, Lee J, Urban A, Malik R, Kang S and Ceder G 2016 Nat. Chem. 8 692 [12] Radin M D, Vinckeviciute J, Seshadri R and der Ven A V 2019 Nature Energy 4 639 [13] House R A et al 2019 Nature 577 502 [14] Saha S et al 2019 Nature Energy 4 977 [15] Lee G-H, Wu J, Kim D, Cho K, Cho M, Yang W and Kang Y M 2020 Angew. Chem. Int. Ed. 59 8681-8688 [16] Vergnet J, Saubanère M, Doublet M-L and Tarascon J-M 2020 Joule 4 420 [17] Yu L, Kokenyesi R S, Keszler D A and Zunger A 2012 Adv. Energy Mater. 3 43 [18] Hinuma Y et al 2016 Nat. Commun. 7 11962 [19] Fabini D H, Koerner M and Seshadri R 2019 Chem. Mater. 31 1561 [20] Castelli I E, Landis D D, Thygesen K S, Dahl S, Chorkendorff I, Jaramillo T F and Jacobsen K W 2012 Energy Environ. Sci. 5 15729 [21] Jain A, Castelli I E, Hautier G, Bailey D H and Jacobsen K W 2013 J. Mater. Sci. 48 6519 [22] Faghaninia A, Yu G, Aydemir U, Wood M, Chen W, Rignanese G-M, Jeffrey S, Hautier G and Jain A 2017 Phys. Chem. Chem. Phys. 19 6743 [23] Sparks T D, Gaultois M W, Oliynyk A, Brgoch J and Meredig B 2016 Scr. Mater. 111 10 [24] Ortiz B R, Gorai P, Braden T, Bensen E, Wilson S D, Stevanović V and Toberer E S 2020 ACS Applied Energy Materials 3 2182–91 [25] Harding J and Pyper N 1995 Phil. Mag. Lett. 71 113 [26] Walsh A and Butler K T 2014 Acc. Chem. Res. 47 364 [27] Torrance J B and Metzger R M 1989 Phys. Rev. Lett. 63 1515 [28] Ganose A, Jackson A and Scanlon D 2018 Journal of Open Source Software 3 717 [29] Horton M K, Montoya J H, Liu M and Persson K A 2019 NPJ Comput. Mater. 5 64 [30] Xie Y, Saubanère M and Doublet M-L 2017 Energy & Environmental Science 10 266 [31] Ong S P, Richards W D, Jain A, Hautier G, Kocher M, Cholia S, Gunter D, Chevrier V L, Persson K A and Ceder G 2013 Comput. Mater. Sci. 68 314

6 IOP SciNotes 1 (2020) 024805 D W Davies et al

[32] Ragavendran K, Vasudevan D, Veluchamy A and Emmanuel B 2004 The Journal of Physical Chemistry B 108 16899 [33] Buckeridge J, Butler K T, Catlow C R A, Logsdail A J, Scanlon D O, Shevlin S A, Woodley S M, Sokol A A and Walsh A 2015 Chem. Mater. 27 3844 [34] Kato D et al 2017 JACS 139 18725 [35] Metropolis N, Rosenbluth A W, Rosenbluth M N, Teller A H and Teller E 1953 J. Chem. Phys. 21 1087 [36] Mathew K et al 2017 Comput. Mater. Sci. 139 140 [37] Jain A et al 2015 Concurrency Computation 27 5037 [38] Kresse G and Joubert D 1999 Phys. Rev. B 59 1758 [39] Kresse G and Furthmüller J 1996 Comput. Mater. Sci. 6 15–50 [40] Kresse G and Furthmüller J 1996 Phys. Rev. B 54 11169 [41] Perdew J P, Ruzsinszky A, Csonka G I, Vydrov O A, Scuseria G E, Constantin L A, Zhou X and Burke K 2008 Phys. Rev. Lett. 100 136406 [42] Krukau A V, Vydrov O A, Izmaylov A F and Scuseria G E 2006 J. Chem. Phys. 125 224106